self-improvement-loops
An agent skill by muratcankoylan, from muratcankoylan/Agent-Skills-for-Context-Engineering. Tags: automation, debugging, developer-tools, performance, strategy.
What it does
This skill should be used when the harness, scaffold, workflow, or optimizer itself is the optimization target: recursive self-improvement (RSI) loops, meta-harnesses, self-improving harnesses that mine their own failures and propose bounded edits, evolutionary or population-based search over agent scaffolds, acceptance gates for self-modifying systems, and agentic context evolution where the mechanism that produces context is versioned and evolved. Route governance of a single autonomous loop (locked surfaces, durable logs, rollback, novelty gates, approval boundaries) to harness-engineering, measurement and quality-gate design to evaluation, judge design to advanced-evaluation, and remote sandbox infrastructure to hosted-agents.
Install
With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:
npx skills add muratcankoylan/Agent-Skills-for-Context-Engineering --skill self-improvement-loopsOr copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):
git clone --depth 1 https://github.com/muratcankoylan/Agent-Skills-for-Context-Engineering
cp -r Agent-Skills-for-Context-Engineering/skills/self-improvement-loops ~/.claude/skills/self-improvement-loopsSafety box score
Grade A, minimal risk: 7 of 100 risk points, rated 2026-10-19. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.
This is a documentation-only skill about recursive self-improvement loops, harness evolution, and acceptance gates; it ships no executable code, installer, or network tool. The main risk is conceptual and operational: it gives an agent a framework for self-editing workflows and archives, which can expand how much code and state the agent is allowed to change if downstream systems are wired to use it.
| Category | Level | Why |
|---|---|---|
| Code & shell execution | 0 (none) | No scripts, shell commands, or code-evaluation entrypoints are shipped; the files are plain markdown. |
| Network access | 0 (none) | There is no downloader or sender; only static reference URLs are listed in `references/loop-design-evidence.md`. |
| Filesystem reach | 1 (low) | It explicitly recommends local archive layouts like `search-run/candidates/...` and storing raw traces and lineage, which implies project-local reads and writes. |
| Secrets & credentials | 0 (none) | No tokens, passwords, `.env` files, keychains, or login credentials are referenced. |
| Privilege & persistence | 0 (none) | No sudo, global install, ծառայ service, shell startup, or OS-setting changes are described. |
| Instruction hijack surface | 0 (none) | The text does not tell the agent to ignore its rules or fetch hidden instructions; it mainly describes evaluation boundaries and governance. |
| Irreversible actions | 1 (low) | The skill is explicitly for "recursive self-improvement (RSI) loops" where an agent "edits its own scaffold" and searches over harness code, but it still reserves promotion decisions for humans. |
| Transparency | 0 (none) | Everything provided is readable markdown; there are no binaries, encoded payloads, or installer artifacts. |
Before installing
- Confirm you want a skill whose goal is to let an agent rewrite its own harness or context mechanism.
- Verify downstream automation keeps evaluator code and permissions outside any editable surface before enabling it.
- If you will rely on the cited research claims, recheck `references/loop-design-evidence.md` because it says the numbers are volatile and need revalidation.
Rated 2026-10-19 by an automated review of 2 files from the skill's repository.
Source
- Repository
- muratcankoylan/Agent-Skills-for-Context-Engineering (all skills from this repository)
- Path
- skills/self-improvement-loops/SKILL.md
- Branch
- main
- Updated
- 2026-09-19
Related skills
- context-degradation — This skill should be used for diagnosing and mitigating context degradation: lost-in-middle failures, context poisoning, context clash, context confusion.
- reasoning-trace-optimizer — Debug and optimize AI agents by analyzing reasoning traces, context degradation, tool confusion, instruction drift, repeated task failures.
- context-compression — This skill should be used when long-running agent sessions need context compression, structured summarization, compaction, token-per-task optimization.
- context-optimization — This skill should be used for improving context efficiency: context budgeting, observation masking, prefix or KV-cache strategy, partitioning.
- harness-engineering — This skill should be used when designing autonomous agent harnesses: research loops, evaluation scaffolds, locked and editable surfaces, durable logs.
- latent-briefing — This skill should be used when the user asks to "share memory between agents", "KV cache compaction for multi-agent", "orchestrator worker context".